simplify format_col.default, allowing e.g. vctrs_list_of columns to print well

[MichaelChirico]

Closes #5948

Turns out, this code in support of #2273 (#5224) is not needed -- I don't notice any difference before/after this change and we pass the related test:

data.table/inst/tests/other.Rraw

Lines 220 to 226 in a36caac

	if (loaded[["sf"]]) { #2273
	DT = as.data.table(st_read(system.file("shape/nc.shp", package = "sf"), quiet=TRUE))
	test(15, DT[1:3, .(NAME, FIPS, geometry)], output="Ashe.*-81.4.*Surry.*-80.4")
	dsf = sf::st_as_sf(data.table(x=1:10, y=1:10, s=sample(1:2, 10, TRUE)), coords=1:2)
	test(16, split(dsf, dsf$s), list(`1` = dsf[dsf$s == 1, ], `2` = dsf[dsf$s == 2, ]))
	}

It's possible our testing is just not extensive enough.

[github-actions]

Generated via commit 9e1260f

Download link for the artifact containing the test results: ↓ atime-results.zip

Task	Duration
R setup and installing dependencies	4 minutes and 33 seconds
Installing different package versions	8 minutes and 6 seconds
Running and plotting the test cases	2 minutes and 20 seconds

[MichaelChirico]

This test fails:

registerS3method("format", "foo2130", function(x, ...) rep("All hail foo",length(x)))
test(2130.15, print(DT), output="All hail foo")  # e.g. sf:::format.sfc rather than sf:::format.sfg on each item

My sense is that if we only fail a toy example, we should just break it. In general I am thinking the better solution here is to add format_col and/or format_list_item methods as needed.

I think given the potential for breaking change, it's best to save this PR for 1.18.0.

[aitap]

Let's compare with other data.frames.

• tibble refuses classed lists that don't come from their own packages:

> tibble(a = structure(1, class = 'foo2130'))
# A tibble: 1 × 1
  a
  <foo2130>
1 All hail foo
> tibble(a = structure(list(1), class = 'foo2130'))
Error in `tibble()`:
! All columns in a tibble must be vectors.
✖ Column `a` is a `foo2130` object.
Run `rlang::last_trace()` to see where the error occurred.
> tibble(a = list(structure(1, class = 'foo2130')))
# A tibble: 1 × 1
  a
  <list>
1 <foo2130 [1]>
> tibble(a = 1, b = list(mtcars))
# A tibble: 1 × 2
      a b
  <dbl> <list>
1     1 <df [32 × 11]>
> tibble(a = 1, b = list_of(mtcars))
# A tibble: 1 × 2
      a               b
  <dbl> <list<df[,11]>>
1     1       [32 × 11]

• data.frame is consistent in saying that format(<data.frame>) is whatever format(...) returns for the individual columns, but doesn't support list columns that well. By default list elements are converted to data.frame columns. Constructing one with a list verbatim requires I(), which overrides format. A list inserted manually into an existing data.frame will be formatted using the usual format method:

> data.frame(a = 1, b = list_of(mtcars)) # NB: uses as.data.frame.vctrs_vctr to make the list into a column
# omitted: similar to current data.table behaviour, i.e., formats the whole `mtcars`
> x <- data.frame(a = 1)
> x$b <- list(mtcars)
> x
# same
> data.frame(a = 1, b = I(list(mtcars))) # stores list as is but with format() overriden
  a            b
1 1 c(21, 21....
> data.frame(a = structure(list(1), class = 'foo2130')) # tries to convert a list to data.frame
Error in as.data.frame.default(x[[i]], optional = TRUE, stringsAsFactors = stringsAsFactors) :
  cannot coerce class ‘"foo2130"’ to a data.frame
> traceback()
4: stop(gettextf("cannot coerce class %s to a data.frame", sQuote(deparse(class(x))[1L])),
       domain = NA)
3: as.data.frame.default(x[[i]], optional = TRUE, stringsAsFactors = stringsAsFactors)
2: as.data.frame(x[[i]], optional = TRUE, stringsAsFactors = stringsAsFactors)
1: data.frame(a = structure(list(1), class = "foo2130"))
> x <- data.frame(a = 1)
> x$b <- structure(list(1), class = 'foo2130') # uses format()
> x
  a            b
1 1 All hail foo
> data.frame(a = I(structure(list(1), class = 'foo2130'))) # uses format.AsIs()
  a
1 1

• DataFrame allows classed lists but expands unclassed lists or data.frames into columns. Seems to ignore format methods altogether:

> DataFrame(a = structure(list(1,10), class = 'foo2130'))
DataFrame with 2 rows and 1 column
          a
  <foo2130>
1         1
2        10
> DataFrame(a = list(1,2))
DataFrame with 1 row and 2 columns
       a.X1      a.X2
  <numeric> <numeric>
1         1         2
> DataFrame(a = list(mtcars))
DataFrame with 32 rows and 11 columns
                      a.mpg     a.cyl    a.disp      a.hp    a.drat      a.wt    a.qsec      a.vs      a.am    a.gear
                  <numeric> <numeric> <numeric> <numeric> <numeric> <numeric> <numeric> <numeric> <numeric> <numeric>
Mazda RX4              21.0         6       160       110      3.90     2.620     16.46         0         1         4
Mazda RX4 Wag          21.0         6       160       110      3.90     2.875     17.02         0         1         4
# omitted
> DataFrame(a = structure(1, class = 'foo2130'))
DataFrame with 1 row and 1 column
          a
  <foo2130>
1         1

• data.table formats lists specially by default, but defers to a format method if it exists:

> data.table(a = 1, b = list(mtcars))
       a                   b
   <num>              <list>
1:     1 <data.frame[32x11]>
> data.table(a = 1, b = list_of(mtcars))
# omitted: formats whole `mtcars` into a single string

Since we allow list columns of any class without applying as.data.(frame|table) first, we might as well format all lists, even classed ones, in a special compact form, despite a few methods, such as sf:::format.sfc, exist to do the right thing. And the sf classes still do print specially because data.table:::format_list_item.default still checks for a format method and eventually calls sf:::format.sfg.

We could also register format_col.sfc and set Enhances: sf. No need for conditional registration here since we own the generic.

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 98.59%. Comparing base (5c964b3) to head (9e1260f).

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #6637      +/-   ##
==========================================
- Coverage   98.59%   98.59%   -0.01%     
==========================================
  Files          79       79              
  Lines       14661    14659       -2     
==========================================
- Hits        14455    14453       -2     
  Misses        206      206              

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.